Research map / Label-efficient and robust learning / Knowledge distillation
Multi-Level Logit Distillation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use multi-level logit alignment instead of vanilla logit distillation | Knowledge distillation · CIFAR-100, homogenous; average across six teacher-student pairs | Top-1 accuracy: 73.09 -> 75.09 | Helped |
| Use multi-level logit alignment instead of vanilla logit distillation | Knowledge distillation · CIFAR-100, heterogeneous; average across five teacher-student pairs | Top-1 accuracy: 71.60 -> 74.93 | Helped |
| Use multi-level logit alignment instead of vanilla logit distillation | Knowledge distillation · ImageNet, ResNet34 → ResNet18 | Top-1 accuracy: 70.66 -> 71.90 | Helped |
| Use multi-level logit alignment instead of vanilla logit distillation | Knowledge distillation · MS-COCO, ResNet101 → ResNet18 | AP: 33.97 -> 36.03 | Helped |
| Add batch-level alignment to instance-level alignment (ablation) | Multi-level logit distillation · CIFAR-100, ResNet32×4 → ResNet8×4 | Top-1 accuracy: 73.33 -> 74.58 | Helped |
| Add class-level alignment to instance and batch alignment (ablation) | Multi-level logit distillation · CIFAR-100, ResNet32×4 → ResNet8×4 | Top-1 accuracy: 74.58 -> 76.26 | Helped |
| Add prediction augmentation using multiple temperatures (ablation) | Multi-level logit distillation · CIFAR-100, ResNet32×4 → ResNet8×4 | Top-1 accuracy: 76.26 -> 77.08 | Helped |
| Vary the median temperature across the reported settings (ablation) | Multi-level logit distillation · CIFAR-100, homogenous | Top-1 accuracy: 74.89; 75.09; 74.92; 74.78; 74.70 | No clear effect |
| Combine the method with RKD | RKD · CIFAR-100, homogenous; average across six teacher-student pairs | Top-1 accuracy: 71.73 to 75.32 | Helped |
| Combine the method with ReviewKD | ReviewKD · CIFAR-100, homogenous; average across six teacher-student pairs | Top-1 accuracy: 74.58 to 75.80 | Helped |
Closest papers by meaning
- Scaled Decoupled Distillation (CVPR 2024 · similarity 0.90)
- Logit Standardization in Knowledge Distillation (CVPR 2024 · similarity 0.90)
- Maintaining Fairness in Logit-based Knowledge Distillation for Class-Incremental Learning (AAAI 2025 · similarity 0.90)
- Enhancing Logits Distillation with Plug&Play Kendall's $\tau$ Ranking Loss (ICML 2025 · similarity 0.89)
- PLD: A Choice-Theoretic List-Wise Knowledge Distillation (NeurIPS 2025 · similarity 0.88)
- Knowledge Distillation with Auxiliary Variable (ICML 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.